I’ve mentioned on the blog several times the continuum that exists between handheld computers and pocket computers, battery powered devices in rather small form factors that are nevertheless fully-fledged general purpose computers — arguably more so than the modern locked-down smartphone has become. Some of these diminutive systems are best considered "handhelds," with larger size, larger keyboards, more power and (often) less battery life, and some are definitely "pocket computers," with smaller size, smaller keys, less power and (usually) better battery life. For example, systems like the Tandy PC-4/Casio PB-100 or Tandy PC-3/Sharp PC-1250 would be considered "definitely a pocket computer," while the Epson HX-20 or Kyotronic 85 systems like the NEC PC-8201A or TRS-80 Model 100 would be considered "definitely a handheld computer," and you can probably think up some examples in between.

Well, here’s a notable example of one single architecture that birthed both types of machine, and it came from a company not really noted for either one: Texas Instruments.

TI certainly made calculators and many of those were programmable by some means, but neither handheld computers nor pocket computers had categorically been in their repertoire to date. Nevertheless, here we have the 1983 Compact Computer 40 — using the AA battery for scale, at that size definitely a handheld — and above it the 1985 TI-74 BASICALC, notionally a "BASIC programmable calculator," but actually an evolved version of nearly the same hardware in less than half the size. Thanks to the ingenuity of the Hexbus interface, which due to TI’s shortsightedness was never effectively exploited during that era, we can get a serial port running on both of these with hobbyist hardware. If we have a serial port, that means we can bring up a terminal program — which we’ll write from scratch in assembly language for shell and Gopherspace access.

But how would a Unix shell work on a single line screen, or for that matter, a Gopher menu? We’ll explore some concepts, but before we do that, for context and understanding of their capabilities we’ll start with the history of these machines — and because their development is unavoidably tangled with TI’s other consumer products and their home computer family, we’ll necessarily rehash some of those highlights and nadirs as well.

The Compact Computer family and the CC-40

The obvious progenitors of the pocket-handheld computer family were the programmable calculators emerging from various manufacturers in the mid-1970s. In this genre TI was a notable vendor despite being just one of several. Arguably the first of these was the Hewlett-Packard HP-65, released in 1974. Although not a computer in the conventional sense (despite HP’s marketing), it could accept up to 100 lines of "program text" (key presses) and read and write them using magnetic stripe cards, and was robust enough to go into space with the Apollo mission in 1975. TI fired back in 1975 with the SR-52, also programmable with mag stripe, and supporting 224 lines.

HP, TI and Casio became probably the most active in this submarket, culminating in the 1977 TI-58 and TI-59 supporting up to 5000 lines of program text stored in "solid state" ROM modules that acted like library cartridges (the ad here is from a 1979 promotion). These modules became very popular for vertical markets, with bespoke program functions stored ready-to-go for industries like insurance and finance and engineering not unlike the later Panasonic HHC. Having knocked other smaller companies like Commodore out of contention, TI understood HP to be their most direct competitor at that time, and launched what was called Project X in 1977 to leapfrog them in the programmable calculator segment: wider screens up to 16 characters wide, expandable memory and an "Equation Operating System" implementing a scrolling display for formula entry and editing, facilitated using a new system architecture tentatively named "LCD III."

As competing calculators added power and new capabilities, a higher end model dubbed the Advanced Language Calculator (or ALC for short, codenamed "Lonestar") sought to introduce more computer-like features to the handheld range with its own scrolling LCD and cartridge-based software and programming languages. The idea turned out to be an atypically forward-thinking move on TI’s part because the market dramatically changed in 1980 when Sharp introduced the PC-1210 and PC-1211, the first true pocket computers with full alphanumerics and programmability in BASIC (the PC-1211 was rebadged by Tandy Radio Shack as the PC-1). In quick succession followed the Panasonic HHC family, even based on a more conventional architecture, the 6502, and shortly afterwards Casio introduced their own pocket line.

This new offshoot of the calculator market caused TI management to reexamine how Project X and the ALC should be positioned. Commissioning a user survey in July 1981, TI’s marketing team broadened both concepts into three potential niches (RM 1000, RM 2000 and RM 3000), reworking the original programmable calculator product into the lowest price point (1000) while adding a BASIC-based single-line LCD pocket computer (2000) and a full multi-line LCD portable as higher tiers (3000). The obvious inheritor of Project X, the RM 1000 eventually morphed into the still-born TI-88 calculator, produced and prototyped but unexpectedly cancelled pre-launch in 1982 due to fears it was already obsolete.

The RM 2000, on the other hand, wasn’t an exact physical match for the ALC/Lonestar, which by then was a larger handheld with a 31-character LCD and "Chiclet" style keyboard, but Lonestar was sufficiently far along in development that its present form factor won out. (The scan above of an undated ALC mockup came from a bad photocopy tucked in with one of my CC-40s which I have tried to airbrush and clean up for legibility. I don’t know anything about its provenance otherwise.) Additional market separation was achieved by using a new 8-bit processor design, admittedly with less capability than their flagship 16-bit 9900/9995 CPUs and microcontroller variants, but also without their legacy baggage. Implementing an internal architecture not unlike the Intel 8051, with which it was intended to be comparable, the TMS7000 series was also positioned as a follow-on to their very successful 4-bit TMS1000 family in that custom microcode was fully supported like the earlier product. Like most microcontrollers, it could be shipped with built-in mask ROM directly from manufacture. TI started selling NMOS variants of the chip family in 1981, but the then-current version of the chip required two nine-volt batteries in the Lonestar prototype to power it and engineers proceeded converting the chip family to low-power CMOS for the ALC.

In parallel, the Texas Instruments consumer products division was determined to maximize sales volume, initially through discounts, reductions in production cost and heavily courting retailers, and in September 1982 newly promoted division head William Turner openly ignited a home computer price war with Jack Tramiel’s resurgent Commodore. Turner aimed to push Commodore out of the market again by competing on price, now Tramiel’s home turf, and use the installed base to move their vendor-locked software and peripherals. While TI sales duly increased, Commodore persistently hung on in the low end market as in retrospect the VIC-20 was a much cheaper system against which the expensive and arguably overengineered TI 99/4A could not indefinitely compete. Nevertheless, through the rest of 1982 the 99/4A outsold the VIC-20 and for some period of time was the number one home computer in the United States, and management rapidly plotted how to diversify the line. Alongside wilder ideas like the 99/7 terminal and a direct successor alternately referred to as the 99/4B or 99/5, at least two concepts got as far as actual hardware. The 99/8 (codenamed "Armadillo") was intended to attack the high side of the market, a substantial expansion of the 99/4A with more RAM and built-in Extended BASIC, while the 99/2 was meant to hit the ultra-low end as a direct response to systems like the UK Sinclair ZX81 (subsequently introduced States-side in slightly modified form as the Timex Sinclair 1000). Although also a full 16-bit member of the 9900 family — with a TMS9995 CPU, the same as the Tomy Tutor — the diminutive 99/2 used a Chiclet style keyboard similar to Lonestar and a very limited black-and-white video chip, less capable than the 9918A but also less expensive. It supported cassette tape and cartridges, and as a stepping stone to the larger systems, BASIC programs written on a 99/2 were designed to be upwardly compatible with the 99/4A and 99/8.

For expansion options, TI created a new and simpler peripheral interface called Hexbus (also variously seen as Hex-Bus, HEXBUS and HexBus in period literature and user notes, though here I’ll render it Hexbus for the remainder of the article). TI positioned Hexbus as a universal means to service all but the highest-bandwidth devices, from modems and serial communications to printers and mass storage. As such, they replaced Lonestar’s original 9-pin I/O port with Hexbus and additionally fitted it to the 99/2 and the 99/8. The protocol is straightforward to understand and implement, and I’ll talk a bit about the technical aspects a little later since the devices we’ll use all connect that way. TI planned to extend Hexbus support to the 99/4A and RM 3000 (in whatever form it ended up taking) so that as many of their comsumer computers as possible could use the same peripherals.

Meanwhile, as Tramiel began driving price cuts faster and deeper and TI was repeatedly forced to match them (as the old Klingon proverb says, revenge is a dish best served cold), management began considering the ALC/Lonestar as a potential anchor in the portables market where competition was not nearly so fierce. The ALC prototype (now the ALC-A, for Product A) was extended into a family of machines: a low-end ALC-LC ("Low Cost") with a 16-character LCD, limited BASIC, Hexbus but no cartridge port, and 1K of RAM; a high-end ALC-C to succeed the RM 3000 with a 40x6 character LCD, larger keyboard, cartridge port, Hexbus, cassette port and 8K of RAM (a terminal variant ALC-T added a 300bps modem); and then the ALC-A itself expanded into the ALC-A2, with the same 31-character LCD and original Chiclet keyboard, plus 2K of RAM, cartridge port and Hexbus. The cartridge port could accommodate programming languages like BASIC, Fortran and Pascal as well as application software, and Turner insisted it also allow RAM expansion.

TI paused the ALC-LC and ALC-C to get the ALC-A2 to market first, reasoning the design could easily be reworked for the others if it was successful. Renaming it as the Compact Computer 40, BASIC was made internal and the original drop-in cartridge port design (shown here on a late prototype) was reworked to a more typical card edge. Additionally, as 2K RAM was increasingly becoming inadequate even for portable systems, TI pumped the base RAM up to 6K — we’ll talk about how TI arrived at that number presently. Now that everything was CMOS internally, it could run sufficiently well on just four AA batteries, which was a significant improvement in convenience, runtime and weight; an optional AC adapter could be connected for desktop use.

The 99/2 and the revised 6K CC-40 made their debut at Winter CES 1983. The "40" came from TI’s specious then-custom of counting ROM and RAM together as total "memory," so according to their vaguely disingenuous marketdroids at the show the CC-40 had 40K of "memory" from its 6K of RAM and 34K of ROM (an unusual but truthful number that again I will explain when we get to the internals). This then carried over to the magazine copy in Creative Computing April 1983, above, where an overly credulous David Ahl (et al) completely swallowed the TI baloney and said it was "expandable to 128K thanks to the 16-bit processor."

In reality both parts of that statement were false: it still ran an 8-bit TMS7000-series CPU, though at the time TI was careful not to say exactly what the CPU was, and TI’s memory claim was more ridiculous than Ahl reported as they actually alleged a ceiling of 168K of "memory" — because it accepted cartridges of up to 128K of ROM (i.e., 6 + 34 + 128). A 16K RAM expansion was also planned and TI estimated a runtime of 200 hours on four alkaline AA batteries.

TI was justifiably very proud of Hexbus and demonstrated at least three peripherals at the show: an RS-232 serial interface for $100 [around $325 in 2025 dollars], a four-colour printer-plotter which many of you have already guessed the mechanism of and selling for $200 [$650], and a licensed version of the Exatron Stringy Floppy that TI reworked into what they called "wafertape." Its sole mass storage option, the wafertape drive was priced at $139 [$460] plus media cost, which wasn’t disclosed. A modem, full-size printer, wand input device and a video interface were announced for later in the year.

Despite being arguably more powerful, the 99/2 was clearly positioned as the junior product — TI priced it at just $100 [$325] while the CC-40 retailed for $249.95 [$810]. Software support likewise reflected the CC-40’s higher priority. TI announced eight cartridges at the show, but only two of them were for the 99/2 at $19.95 [$65] each, and about twenty more promised titles were to be released on cassette tape at $9.95 [$32] a pop. On the other hand, the CC-40 got at least six cartridges (seven listed here) and at least thirteen additional software packages on wafertape, with a promised seventy-five more; the most expensive of these retailed for almost $150 [$485]. TI informed attendees both computers would be available by summer.

Unfortunately, what should have been a triumphant return for TI at Summer CES 1983 turned into disaster. Commodore, fresh off the introduction of the Commodore 64 in 1982 at $595, had slashed the VIC-20 to a new low of $125 in January 1983 and once again forced TI to follow suit. By the time the June convention rolled around, Commodore had managed to gobble up over 50 percent of the home computer market and largely at TI’s expense: the Commodore 64 was flying off shelves at under $300 — half its introductory price, even lower at discounters — and the VIC-20 continued selling briskly for $100 or less. In contrast to the 99/4A, Tramiel’s high degree of vertical integration and low production costs ensured Commodore could still turn a profit. Nevertheless, TI doggedly matched the VIC-20’s new price (with a rebate) and started taking a significant loss on every unit, which also put them in the distinctly unpalatable position of introducing the notionally low-end 99/2 at or above what the 99/4A was now selling for. The 99/2 was quietly cancelled the day before the show and TI later liquidated the small amount of stock that had already been manufactured.

In fairness, TI was not the only one in trouble and both Atari and Mattel took massive losses as well. 1983 was also the year of the crap home computer with many other underwhelming systems being introduced at or near that magic $100 price point — and I know if I mention any by name, someone out there will howl because there will always be at least one person who loved theirs. Unlike TI’s hardware, however, such systems were made cheap to sell cheap: while most customers correctly perceived them as technologically inferior, they at least covered their cost per unit, even though most of these fly-by-night also-rans flamed out within months.

Seeing all this upon arrival in Chicago, TI management abruptly decided not to exhibit the 99/8 either and hastily put the display model behind a locked door for the show’s duration. (Ironically, the Tomy Tutor — nothing less than a modified clone of the 99/8 — made its American debut at the very same convention.) Although a public relations debacle of its own (some mischievous press coverage even reportedly printed pictures of the door), TI’s most enduring announcement from that Summer CES was a threat to completely lock out third-party software for the 99/4A, a deeply unpopular and thinly veiled attempt to maximize its failing razor-and-blades model, and one they mostly made good on by enforcing their patents on GROM chips. The company’s stock cratered after announcing a $119 million loss in July — both due to production costs and a critical overestimation of demand — and Bill Turner was forced out of the company. Incredibly, Turner’s replacement Peter Field decided to double down on the strategy by slashing the price of peripherals as well as the computer.

On the other hand, the CC-40 largely sailed above this fray because it was positioned as a business product. Although ostensibly developed and sold by the same consumer products division — unlike the TI Professional PC from the data systems group which had also come out around this time — marketing took great pains to distance it from the ailing home computer line and emphasize its utility for serious business rather than general purpose entertainment. Bill Cosby, celebrity endorser of both TI since 1982 and Jell-O, which closely resembled the spines of certain TI executives, was there to remind you there was always room for a CC-40 (and prison time).

In the official marketing pamphlets TI provided at Winter CES 1983, you can see the difference in tone between the CC-40’s and the contemporary 99/4A’s: Cosby is more casually dressed for the 99/4A and it appears prominently billed as a home computer in two places just on the cover alone, yet for the CC-40 he presents the computer and a wafertape while wearing business attire, using executive buzzwords like "solutions" and making no mention of home applications at all. (This particular pamphlet is the source of some of the scans I showed you previously.) Based on early hands-on reviews around August 1983 and estimating about a two to three month publishing delay, it appears the CC-40 made it to retailers at the promised price point between May and June of that year.

The released CC-40 is very similar to the ALC concept; the CC-40 retains Lonestar’s general form factor and its 31-character non-dot-addressible LCD with various mode indicators. These indicators not only include things like the state of the SHIFT key and the current trigonometric unit of angle, but also unique user-addressible arrow-like markers along the bottom of the LCD panel which can be programmatically toggled to signal the user. Limited custom graphics were possible by redefining one of six screen characters dedicated to the purpose.

The keyboard, however, was one thing that did change. While the ALC and the CC-40 use the same "Chiclet" keyboard in the same overall layout, including a small numeric keypad, some of the keys were moved around and certain others removed. The undated ALC image I have above likely came from the same source as these documents from a former TI engineer, one of which shows the MODE key crossed out, the LBL key replaced with CTRL and the equals key at the bottom left of the numeric keypad replaced with a generic "FN" function key. It looks like the ALC was also proposed to support some sort of Japanese language entry, most likely katakana (the display was not sufficient for most hiragana, let alone full-blown kanji), but this feature was also subsequently eliminated. On the image here I can’t distinguish what the shifted characters were meant to be, but this photograph appears to be of the same or similar ALC mockup, showing some typographical characters also moved around as well as a line of deep blue secondary functions possibly for debugging purposes.

If you have big hands, then you probably have big thumbs, and you know what they say about guys with big thumbs: they can type better on handheld computers. (What? This is a family blog.) In my clodhopper not-quite-varsity basketball mitts the CC-40 is almost a handheld but it wasn’t for most people, and even more so than its single-line LCD the roughly 80% keycaps — although considered better than the 99/2’s mushy Chiclets — provoked polarized opinions. Joe Devlin in Creative Computing said the keyboard was "not all it might be," but found it "well-built" with "no hint of keybounce that is a frequent plague," while Microcomputing asserted it was "basically a calculator keypad with letters on it" and Byte’s David Ramsey bluntly called it "vile." On the other hand, it was certainly no worse than, say, an HP-75. A small recessed divot next to the space bar acts as a soft reset button; the machine also comes with a translucent keyboard overlay with quick combinations to type BASIC keywords, though this falls off easily and I don’t much use it myself.

The underside of the unit contains the battery door (four AA batteries) and a sturdy flip-up stand for using it on a desk. There are supposed to be four rubber feet but the glue kind of sucks and two of them came off at some point. The CC-40’s official model number, shown here, is the "CC 40." This unit is serial #A001818 with date code ATA3583, indicating it was assembled at TI’s plant in Abilene, TX in the 35th week of 1983 (i.e., end of August).

The CC-40 has just three external ports. On the left side of the computer (right in this image) are the AC power jack and the Hexbus connector, and on the right is the single cartridge port. The power jack, officially TI’s model AC 9201 wallwart, is a standard barrel connector providing 6V DC but atypically negative tip. However, kudos do have to be given to TI for its power implementation: people have reported accidentally connecting a positive tip AC adapter will not let out the magic smoke (it just won’t work, though always try to avoid it), and you can use the AC adapter to bridge changing the batteries while it’s running. The CC-40 will seamlessly switch to AC power when it’s available and batteries when it isn’t, including the situation where the barrel jack is plugged in but the wallwart is not. If all else fails, I have read various reports that the system will maintain memory contents for about a minute if you can swap the batteries quickly, which matches my personal experience. That said, I use lithium AAs in mine and I have yet to replace them.

Unlike the drop-in cartridges of the prototype, released cartridges have a more typical card edge protected by a sliding cap which guides the edge into the slot. The ROM software cartridges and the RAM expanders have the same form factor, but there is only one slot, so it is not possible to have a RAM expansion and a ROM cartridge plugged in together. (Although modern homebrew combo cartridges exist with both ROM and RAM, they don’t support accessing both at the same time.)

As you saw in the extract from the marketing pamphlet, TI called their ROM cartridges "Solid State Software." However, not all the promised cartridges were ever released. TI only officially sold the Statistics, Finance, Pascal, Memo Processor, Games I, Electronics ("Advanced Electrical Engineering") and Mathematics cartridges, though the Editor/Assembler cartridge was subsequently found and dumped, and nowadays appears on most multicarts. On the other hand, despite being officially planned, Business Graphics and Games II have yet to be discovered. These cartridges appear to the computer as programs which can be run by name; some cartridges like Games I have multiple programs (Hunt the Wumpus, Hammurabi, Stocks and Bonds, Spacecraft Landing Simulation and Backgammon; Games II would have had Codebreaker, Interchange, Nim, Sea Battle, Number Gallery and Word Scramble). Some cartridges can also add functionality to BASIC programs. We’ll look at one of these cartridges in particular a little later on.

TI produced two RAM cartridges for the CC-40, though because of the CC-40’s unusual BASIC memory map, not all of their RAM was useable under all circumstances. The most common, and arguably the most useful for a particular reason I’ll mention shortly, was the 8K Constant Memory cartridge. This cartridge contained battery-backed RAM which could either be added to the system as expansion memory for BASIC (but resets the system when taken out), or be used as a removable storage medium with a small machine language program. In this second mode the entire memory of the standard 6K system could be copied to the cartridge and back, and since it carried a small lithium battery, its contents — including the machine language subroutine necessary to load from it — would persist when unplugged. TI estimated this battery would last about three to five years using a mechanism that would connect the battery only on first use. Naturally, unless you find an unused NOS unit, current examples are dead as doornails but despite not being user-serviceable the batteries can be replaced with some effort.

The second option was the 16K RAM cartridge that TI had originally promised. This unit lacked a battery and was completely volatile. Although technically capable of serving as a storage medium, it could only do so while the cartridge remained connected and the computer’s batteries were good, so it was most useful simply as expanded memory. Unfortunately, while the original 2K iteration could take advantage of all 8K and 16K (to yield 10K and 18K respectively), BASIC’s memory map in the released 6K CC-40 couldn’t access 4K of either device — which is to say the 8K cartridge could only add 4K of memory to BASIC to yield 10K (though all 8K was addressable), and the 16K only 12K to yield 18K. Although TI did mention a 2K expander in documentation, no such cartridge was sold. Despite this deficiency, the CC-40’s memory map can support up to 32K of external memory and modern homebrew RAM cartridges typically provide all 32K with battery backup a la the original 8K Constant Memory.

That brings us to the eight-pin Hexbus port, though the manual calls it the "peripheral port." Hexbus is a four-bit parallel interface over an extended bus capable of addressing up to 255 intelligent devices (device 0 is "all devices"), and the "hex" part likely reflects that it can send an entire hexadecimal digit at once with the remainder for ground, handshaking and future expansion. It is hot-pluggable and largely device-agnostic, and like the analogous Atari SIO system can be considered a spiritual ancestor of modern USB. Hexbus hosts are generally the centre of the bus (unless this role has been reassigned) and as such order downstream devices to act or report using command and data packets. It is possible to interrupt the host when a device requires service, though this requires the host to subsequently poll each enabled device to determined who interrupted it. (Foreshadowing.) Power is not distributed over Hexbus sensu stricto but we’ll get to a later variation that can.

TI intended Hexbus to service all kinds of peripherals and assigned fixed device number ranges to various classes. In practice this was not always strictly observed, nor were all assigned device classes actually used or all possible numerical ranges actually assigned. Among others, officially 1-2 were for cassette tape but 3-8 for tape mass storage, 10-19 for various sorts of printers (with subclasses beneath them) but 50-53 for Centronics-style parallel interfaces, 20-23 for RS-232 serial and 70-73 for modems, 30-33 and 40-43 for colour and black and white video interfaces, 100-109 for 5.25" floppy disk drives, 110-117 for 3.5", and 90-95 for barcode readers and scanners. 118-239 was explicitly dedicated to third-party use but most homebrew Hexbus devices, including the ones we’ll use in this article, emulate the standard device IDs. Thirty-one different commands are defined; devices are only required to implement those which are relevant to their operation.

Hexbus mandated the host be capable of no less than 3000 bytes/second for "peripherals that require a minimum I/O transfer rate" and the CC-40’s user’s guide rates the CC-40 as capable of up to 6000 bytes/second. Theoretically a nybble could be transmitted about every 16 microseconds, yielding a maximum throughput of around 31,250 bytes/second, though rarely was anything close to this top speed realized. Part of the problem was the protocol’s overhead as well as the computers themselves, which for reasons of power consumption did not drive the bus particularly efficiently under ordinary circumstances. The other wart on Hexbus are the cables, which have poor strain relief and rely on a sometimes mushy friction fit in the ports, which themselves aren’t well-reinforced either.

As mentioned, the launch portfolio for Hexbus was a printer/plotter, RS-232 serial interface and the wafertape drive, and this three-pack was what appeared in the 1983 CC-40 pamphlet. For the 99/4A’s 1983 pamphlet, however, notice that not only were Hexbus devices explicitly advertised with the new beige 4A (along with a beige PEB which I’ve never actually seen and was probably a mockup), but there were now four of them.

Those four are shown here: the small printer/plotter (HX-1000 Printer/Plotter), the RS-232 box (HX-3000 RS-232), the wafertape (HX-2000, though this one is unmarked) and additionally a modem (HX-3100 Modem). They were intentionally stackable into an attractive cube configuration, though I note the printer/plotter has a different front bezel colour that doesn’t match the other devices or what appeared in the pamphlets. Although the modem and wafertape can accept regular batteries, the printer/plotter is powered by a NiCad pack with all the attendant disadvantages and uses its own charger (AC 9203, nominally 9V center-tip but my multimeter reads about 12.5V), and the RS-232 box must be connected to wall power (using another AC 9201 adapter, same as the CC-40 itself).

Anyone who knows anything about plotters of this era will recognize the mechanism in the printer/plotter: it’s another Alps DPG-1302, the same as the Tandy CGP-115, Commodore 1520, Atari 1020, Oric MCP-40 and the Convergent Microprinter WP-100, among many others. However, that also means that refurbishing it is fortunately quite practical once you replace that hideous internal battery pack.

Later on a full-size 80-column dot matrix printer (HX-1010 Printer 80) became available, though unlike the plotter this one can take regular D batteries (four). This printer is based on a Brother printer mechanism also found in their EP-20 electronic typewriter and uses the same ribbons, or if you have no ribbons, you can use it as a thermal printer with fax paper such as that for the Silent 700 series. I don’t have the AC 9401 supply the Printer 80 uses, but based on the EP-20 (which also could run on four D batteries) it is likely also 6V tip negative at 1.0A, unlike the 300mA AC 9201.

More so than the others, however, the wafertape drive was intended to be the star of the show. It is in fact a portable version of the Exatron Stringy Floppy (from Exatron in Sunnyvale, later Entrepo, Inc.), a high-speed (for the era) continuous-loop tape system first introduced in 1978 for which Texas Instruments had negotiated an exclusive license in 1982. Like other forms of tape storage the Stringy Floppy is inherently sequential, and has no true provision for random access — files must be scanned through in order until the desired one is encountered. But, because it’s looped and streams fast, it can nevertheless achieve speeds comparable to a conventional floppy disk system. The wafertapes ("microwafers") fit easily in one’s hand, or at least they do in my hand.

The length of the tape naturally determines its capacity, although longer tapes are also unavoidably slower to search. In my collection I have 5 foot, 10 foot, 25 foot and 50 foot wafers, apparently the only sizes TI offered, which in a somewhat non-linear fashion respectively correspond to maximum capacities of 4K, 12K, 32K and 64K. The wafertape drive is typically device 1, the same range as cassette tapes, though a selector in the back can pick a device ID anywhere from 1 to 8 if you were crazy enough to have a stack of them.

In retrospect TI’s choice of the Stringy Floppy was peculiar because the device already had a somewhat chequered reputation on other platforms. Although reasonably fast, its storage capacities were unimpressive compared to floppies and the media wasn’t cheap. Worse still, the tape media was well-known for being prone to tangling and even sometimes outright fracturing, a problem that only got worse as tapes got older and more fragile, and tended to happen more with expensive higher-capacity cartridges because the tape in them was thinner. Repairing a mangled or broken tape could be an exceptionally fiddly business which I’ve experienced enough firsthand to the point where I rarely use the wafertape drive anymore except for demonstrations.

Yet all that was not sufficient to deter TI, who wanted to avoid the association of ordinary cassette storage with home computers, and believed the wafertape drive would be seen as handy, speedy and state of the art. However, two more flaws subsequently became apparent. The first was that tape handling became even more unreliable on battery power, especially as the batteries got low, and virtually necessitated the device be connected to the wall. But the second, discovered in October 1983, was a rare but critical fault in its internal microcontroller where under specific circumstances it could unexpectedly fail to wake on an interrupt and thus go out of synchronization.

By the time this fault was discovered, CC-40 owners were already howling because of the backlog on peripherals — any peripherals. These didn’t start hitting retail channels until the fall and sometimes even later: my modem has an ATA3483 date code while my RS-232 box has an LTA4983 date code, well into December, and my NOS printer/plotter has a date code of 014 indicating January of 1984. The delays were significant enough that as early as July TI had already announced expected declines in CC-40 sales, at the same time Bill Turner got cashiered, and ended up taking writeoffs. Although the wafertape drives were being manufactured on the same timetable (mine has an ATA3383 date), at that point the CC-40 was clearly in decline and TI management didn’t want an expensive recall or lawsuit incited by the drives’ flaws on a product entering twilight. The wafertape drive was cancelled before it was ever sold and stickers saying so went out on every CC-40’s box thereafter. (Note the box copy: the copyright says 1982, and the modem wasn’t ready by then, so it doesn’t appear.)

But since there were stacks of other boxes and documentation already printed referencing the wafertape, TI wasn’t content to mark just the CC-40’s box alone with its discontinuation; they also jammed slips in manuals and slapped stickers on the boxes of the other peripherals so that they didn’t need to be reprinted (and no one could claim TI didn’t warn them). This ended up making the 8K Constant Memory cartridge practically essential because without it you had no chance of saving anything.

One other mass storage device was created by TI for Hexbus, a 5.25" floppy drive, but while it may have been able to work with the CC-40 this device was actually intended for the home computer line and went unreleased. There was also a prototype Hexbus colour video interface, in practice more like a serial terminal that accepts commands from the CC-40 and draws a screen with its own internal 9918 (actually a 9118, which is functionally equivalent). The also-unreleased Editor/Assembler cartridge can control this interface for full-screen editing. Without a viable mass storage option, however, there didn’t seem much point in releasing either one.

Meanwhile, TI had always intended to produce the CC-40 in different configurations, sort of a further expansion of the ALC family. In memory configurations alone TI envisioned five buckets at 2K, 4K, 6K, 10K and 18K (and these even appear in the user’s guide), but eventually only publicly produced the 6K and top-spec 18K models. The 18K’s arrival was tardy and it received little fanfare and even less mention in the press.

It surely didn’t help that the only indication it had three times the base model’s memory were two grotty fluorescent green "18K" stickers, like you’d find on a cheapo semi-gold earring stud, and which have peeled off most units since. This is my NOS 18K showing everything that came in the box with every CC-40, including a keyboard overlay (installed), READ THIS FIRST! card, quick reference card, user’s guide (with wafertape warning), and a set of AA batteries. Nowhere else is its expanded capacity mentioned.

You couldn’t even tell from the bottom of the unit; the model numbers (both "CC 40") are exactly the same, and the serial number and date code on this 18K (lower left) are #A000685 and ATA3483, manufactured nearly simultaneously with my usual 6K. On the one hand, the 18K variant was definitely worth getting if you needed the memory: not only did it have more at baseline, but unlike the 6K unit its BASIC could also use the entirety of the 8K and 16K cartridges for up to 34K RAM total. On the other hand, it was obviously impossible to store all 18K in an 8K Constant Memory cartridge, it was never available in large numbers, and most people at the time didn’t even seem to know it existed. Despite being brought in at the same $249.95 price point or less to flog sales, it made little impact.

The reason TI figured they could churn out a whole bunch of configurations is because internally the CC-40 is very simple, just a handful of chips and through-hole discrete components on two PCBs. Here I’m showing the main CPU board so I don’t have to play 69-key pickup with the keyboard/LCD board beneath it. That board is even simpler with just a bog-standard Hitachi HD44780 LCD driver/controller supplemented by an HD44100 LCD driver to handle the additional display segments. The character segments are divided roughly into quarters with the HD44780 handling the common lines and first and third quarters and the HD44100 handling the second and fourth, plus the display indicators.

Orienting the battery compartment to the bottom, the topmost three chips here are the RAM. In the base model these are three 2K 6116 static RAM chips, yielding an unusual total of 6K. However, notice that the two at the top of this picture sit on pads bigger than they are and the one below them does not, plus wire jumpers just above the 5.000MHz crystal oscillator. We’ll come back to those items in a moment. The power and reset circuitry is on the left side set off by a weird little plastic sleeve containing a bridging resistor.

Zooming in a little, we see the two primary chips, the CPU (the DIP on the lower right) and the control ASIC (the QFP). The ASIC was fabbed by American Microsystems (AMI, later AMI Semiconductor, and subsequently acquired by onsemi in 2008) and acts as a "Grand Central" address decoder for the LCD, piezo beeper, internal ROM, cartridge ROM and RAM. It also handles the Hexbus lines and uses the crystal for its own clock source which it additionally passes through to the CPU.

The CPU is a TMS70C20 and indeed a CMOS version of the 8-bit TMS7000 microcontroller we discussed earlier, with the "2" indicating a built-in 2K of mask ROM (the version of the mask ROM is C11002, "C" meaning standard microcode), and a date code of 10th week 1983. The full model designation of "TMX70C20N2L" is interesting for several reasons: "TMX" indicates the earliest stage of production device where the die may not be fully developed (versus the TMP "prototype" where it is, and the TMS "sale" version designation is fully qualified for production), and the "N2" is a 70mil (0.07" or 1.78mm)-spaced DIP as opposed to the more typical "N" which is a 100mil (0.1" or 2.54mm)-spaced DIP, both rated for operation in typical temperature ranges ("L"). It divides down the 5MHz signal from the ASIC to yield its nominal clock speed of 2.5MHz, and has a 16-bit address bus for a maximum 64K addressing space.

TI fabbed the TMS7000 family using what it called "Strip Chip Architecture Topology." Setting out explicitly to make an easily reconfigurable design, TI stepped away from more parsimonious layouts like the 6502’s and platted out its logic in quadrilateral functional blocks it termed "strips." Unlike those earlier designs, and unlike later multi-chip modules using multiple dies, these blocks could be connected with a minimum of additional logic and then promptly unified on a single die for fabrication. One strip included the timer, I/O control, interrupts, on-chip registers and ALU; another held the microcode and PLA for instruction decoding; another held the onboard ROM; another the onboard RAM, and so forth.

You can see one result in the die picture on the front of TI’s product circular. The large strips for the microcode and onboard ROM are easily distinguished (the "darker" strips along the left and centre) as is the on-board RAM (the patterned strip along the right). The ALU/registers/etc. strip is between the two ROMs, and the remainder is smaller functional blocks and glue circuitry. Like the TMS1000 it was intended to succeed, TMS7000s shipped from the factory with optional mask ROM (2K in the TMS702x/C2x, 4K in the TMS704x/C4x and 12K in the TMS70120) and optional custom instruction microcode ("L" series mask ROM designators). To make a different chip, say, with more ROM or a different instruction decoder, a different strip was loaded to replace it — potentially even a different sized strip — while the others could remain the same. (TI’s brochure even brags that the 7040 was created from the 7020 without redesigning the chip: the "design was separated at the memory border and the additional 2K of memory simply inserted alongside the original 2K of memory by the design computer.") The baseline TMS7000 had no ROM, but the development SE70P16x chips carried a socket on top for piggybacking an EPROM (or you could use the TMS7742 which had an EPROM of its own inside), configurations also offered by other microcontroller manufacturers of the period. TI reserved the first six bytes of the ROM for internal purposes. It is possible to dump and save the contents of the internal ROM.

Like the Intel 8051 and Motorola 6801, with which it was intended to directly compete, the TMS7000 family included onboard RAM and I/O standard. The 70x0 and 70Cx0 chips accepted up to a 5MHz clock source (divided by two or four) and provided 128 bytes of onboard RAM, a single 13-bit timer and 16, 8 and 8 (respectively) bidirectional, input and output I/O lines. In the CC-40, those I/O lines are mostly dedicated to the keyboard and memory addressing. The higher grade 70x2 and 70Cx2 chips accepted up to an 8MHz clock source and provided 256 bytes of onboard RAM, two 13-bit and one 10-bit timers, an on-chip serial port and 22, 2 and 8 bidrectional, input and output I/O lines. On the other hand, the TMS7000 was more like the 6801 in that it was a Von Neumann architecture and incorporated the on-chip features as memory-mapped I/O. 10MHz crystals were later supported, though these were still divided down for the internal clock.

The TMS7000 turned out to be a very bright spot for TI; it was substantially faster than the 8051 at most operations and quicker than even an equivalently clocked 6801 or Zilog Z8. I’ll have more to say about the CPU when we get to actually writing assembly code for it.

Here is the ROM (apart from what’s in the 70C20) and the other two RAM chips. This external 32K 61256 ROM stores BASIC and higher level portions of the operating system, while the lowest level portions are within the CPU’s mask ROM strip. Since 32K would consume half the CPU’s addressing space, it is instead banked in using 8K segments (more on that later). Either way that gets you your 34K total of ROM, and in the mildly mendacious world of TI six plus thirty-four equals forty, so that’s the CC-40.

The lower of the two RAM chips has no extra pads. This is in fact the original 2K configuration and what would have been present on any CC-40 variant. This lowest 2K is reserved by the system and not available for BASIC programs in any CC-40 configuration, but can be used by machine language programs and is legitimately addressable RAM.

For the other configurations we crack open the 18K version. The jumper wires I pointed out have moved, and the two upper 2K 6116 RAM chips have been replaced with 8K 6264s (2 + 8 + 8 = 18) which use the extra pads for their additional addressing lines. With this basic template in mind, the other configurations would probably have been no additional chips (2K), two 1K chips (4K), and two 4K chips (10K). It wasn’t long before existing CC-40 owners discovered they could do such an upgrade themselves, to TI’s great chagrin, even though this would clearly void the warranty.

One other notable landmark on the board is this bundle of resistors sprouting from a blob of solder connected to the ground plane. These are pull-downs for the Hexbus lines.

The peripherals necessarily also have their own CPUs, mostly TMS70C20s as well, such as the RS-232 box’s board shown here. This is also a 70C20N2L with the narrower pin spacing and has a ROM code of C14018 with a date code of 10th week 1983. It runs slightly slower at 2MHz from a 4MHz crystal, but instead of bitbanging the serial port it has a real 6551 ACIA made by Synertek with a date code of 19th week 1983. The ACIA has its own 1.8432MHz crystal. A set of headers above the CPU allows the installation of an optional parallel port.

The other notable chip on this board is directly above the SY6551, a 22-pin TI 1052911 manufactured 18th week 1983. This is the Intelligent Peripheral Bus Controller (IBC for short), managing the Hexbus lines for the CPU instead of requiring additional bitbanging, and generating an interrupt when data is available. In later devices it is superseded by the 28-pin TP0370. Everything else is off-the-shelf. We’ll have more to say about the RS-232 box when we get to discussing Hexbus communication options in detail.

The lack of a viable mass-storage option severely damaged TI’s attempts to portray the CC-40 as a useful portable computer. Consequently (and ironically), many owners tended to treat it as an overgrown programmable calculator instead, a task at which it nevertheless excelled in part due to its higher-precision base 100 BCD floating point and polynomial transcendentals, as well as the software on the limited selection of Solid State Cartridges which (although expensive and few in number) were generally high quality. On top of that, it was also substantially faster than TI’s other programmable calculators and was of course capable of more sophisticated algorithms. Some users were able to source remaindered wafertape drives through unofficial channels but a third-party Hexbus storage device, the Mechatronics QD-01 Quickdisk using an unusual 2.8" medium, did not become widely available until 1986. Meanwhile, the consumer products division, now under half a billion dollars in cumulative debt from their ruinous price war with Commodore Business Machines, threw in the towel and announced on October 28, 1983 that they were abandoning the home computer market. The 99/8, all but cut from life support already, was officially canned earlier in October after about 150 had been made; production also immediately ceased for the 99/4A, after which it was heavily discounted and the remaining stock liquidated in March 1984.

Once again, however, the CC-40 temporarily avoided this fate because TI continued to sell and market it as a business product, just like the TI Professional PC, which TI claimed was still doing well. By early 1984 the 6K CC-40 had dropped below $200 at retail, although the 18K version was all but nonexistent in sales channels. TI’s later